LM Studio
by Element Labs
Getting-started guide
Run an AI model entirely on your own computer, through an interface that does not require a terminal.
Current prices: Free · See all plans
Last reviewed 2026-07-27 · for LM Studio
LM Studio is a desktop app that downloads AI models onto your machine and runs them there. You browse a built-in catalogue, download a model file, load it, and chat with it in a window that looks like any other assistant — except the processing happens on your own hardware.
A desktop app for Windows, macOS, and Linux that runs open models locally with a graphical interface.
The point-and-click counterpart to command-line local runners — same underlying capability, no terminal required.
Free for personal and commercial use, with an optional paid tier for organisations needing central administration.
Completely offline once a model is downloaded, which is the reason most people choose it.
Your hardware is the limit, not your plan. Memory is the binding constraint — roughly 8GB of RAM will run small models slowly, 16GB is workable, and 32GB or more with a dedicated graphics card is where it becomes comfortable.
Model files are large. Expect several gigabytes per model and plan disk space accordingly.
Since 2025-07-08 the app has been free for use at work with no licence request, no form, and no fee. Older articles saying otherwise are out of date.
Local models are genuinely less capable than the best hosted ones. You are trading capability for privacy, offline access, and zero per-use cost.
Nothing you type is sent anywhere. That is the feature.
Official source: LM Studio official site
Download the installer from the official site for your operating system. Model files are large — do this on a connection you are not paying by the gigabyte for.
Use the built-in catalogue rather than hunting for files elsewhere. It shows estimated memory requirements before you download, which is the number that decides whether a model will run at all.
Start with a small model. A small model that runs is more useful than a large one that swaps to disk and takes minutes per reply.
If a model loads but responds unusably slowly, it is too large for your machine. Unload it and step down a size rather than waiting it out.
This is the strongest privacy posture available in this catalogue, and it is structural rather than contractual. Your conversations never leave your computer, so there is no retention policy to read, no training toggle to set, and no terms that can change next year and quietly alter what happens to your data.
Worth stating plainly because it inverts the usual advice: with a local model, there is no vendor to trust and no policy to check. The trade is that you also have no vendor to fall back on — no uptime guarantee, no support, and capability bounded by the machine on your desk.
Picks a small model from the built-in catalogue, checks its estimated memory figure against your RAM, downloads it, loads it, and sends one message.
Why this one: The most common first experience is downloading something too large and concluding local models are useless. Establish a working baseline instead.
Tweak: Note the estimated memory figure before downloading. If it is close to your total RAM, choose smaller.
[Paste the document or question you have been avoiding putting into a cloud tool, and ask your real question.]
Why this one: This is the entire value proposition. Feeling the difference makes it concrete.
Tweak: Then disconnect from the internet and ask again. Watching it work offline is the moment it clicks.
Loads a second model of a different size or family, asks it the same question, and puts the two answers side by side.
Why this one: Teaches that "an AI model" is not one thing, and calibrates your expectations for what a machine your size can do.
Tweak: Use a question you already know the answer to. You are evaluating the model, not learning the topic.
Lets the model read a file from your machine without uploading it anywhere.
Why this one: Private document analysis is the strongest practical use case for a local setup.
Tweak: Start with a short document. Long ones are where limited hardware shows most.
Want the full picture first? See the LM Studio overview for pricing, the free tier and related tools.
